MLA-C01 Question 52
Select 3You are building a machine learning pipeline on AWS to predict customer churn. Your dataset has several missing values and contains duplicate records. Additionally, some numeric features have extreme outliers that significantly skew the data distribution. Which of the following are appropriate data cleaning and transformation techniques to prepare your data for modeling?
- A
Use the AWS Glue FindMatches transform to deduplicate records in the dataset.
- B
Replace missing values with the mean or median of the respective columns.
- C
Remove all rows containing missing values to ensure data consistency.
- D
Use robust scaling methods such as the interquartile range (IQR) to reduce the impact of outliers.
- E
Apply the PCA (Principal Component Analysis) algorithm to handle missing values.
Show answer and explanation
Correct answers: A, B, D
Explanation
Data cleaning and transformation are critical steps in preparing a dataset for machine learning. AWS Glue FindMatches can deduplicate records effectively. Missing data can be handled through imputation methods like replacing with mean or median values, which preserves the dataset's integrity. Outliers are best managed by robust scaling methods such as IQR, which limit their impact without discarding them. Removing rows with missing data or using PCA for handling missing values are not ideal approaches in this scenario.
- A. Correct.
Correct. AWS Glue FindMatches is designed to identify and deduplicate similar records in datasets, making it a suitable solution for handling duplicate data.
- B. Correct.
Correct. Replacing missing values with the mean or median is a common and effective imputation technique to ensure that the data remains usable without introducing biases.
- C. Incorrect.
Incorrect. Removing all rows with missing values can lead to significant data loss, especially when the dataset has many missing values. This approach is generally not recommended unless the missing data is minimal.
- D. Correct.
Correct. Robust scaling methods like the interquartile range (IQR) are effective in mitigating the impact of extreme outliers without removing them, ensuring the data remains reliable for modeling.
- E. Incorrect.
Incorrect. PCA is a dimensionality reduction technique, not a method for handling missing values or cleaning data. It is unrelated to addressing missing data or outliers.